What I built
Link mobile-device locations to population data, then estimate segregation by hour.
- 01Infer home
Cluster nighttime locations to infer home areas, then link each area to Census population characteristics.
- 02Reweight the sample
Align the device sample with Census population composition.
- 03Compute at scale
Calculate distance weights and the multigroup Theil segregation measure in compiled C++ called from R.
- 04Aggregate by hour
Estimate segregation for 428 hourly intervals in each of 18 commuting zones—regional labor markets defined by commuting patterns.
Compared with residential segregation
Daily mobility reduced segregation relative to residential measures in most regions.
Residential segregation uses where people live; the hourly estimates incorporate where people spend time throughout the day.
In 13 of 18 commuting zones, even the highest dynamic-segregation estimate remained below the residential benchmark.
